Skip to contentSkip to conversation

mtd: nand: write BBM to OOB even with flash-based BBT

PATCH Applied Brian Norris 3 participants 3 messages Feb 6, 2012 – Feb 8, 2012

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v5 1/1 mtd: nand: write BBM to OOB even with flash-based BBT

Currently, the flash-based BBT implementation writes bad block data only to its flash-based table and not to the OOB marker area. Then, as new bad blocks are marked over time, the OOB markers become incomplete and the flash-based table becomes the only source of current bad block information. This becomes an obvious problem when, for example:

* bootloader cannot read the flash-based BBT format
* BBT is corrupted and the flash must be rescanned for bad blocks; we want to remember bad blocks that were marked from Linux

So to keep the bad block markers in sync with the flash-based BBT, this patch changes the default so that we write bad block markers to the proper OOB area on each block in addition to flash-based BBT. Comments are updated, expanded, and/or relocated as necessary.

The new flash-based BBT procedure for marking bad blocks:
(1) erase the affected block, to allow OOB marker to be written cleanly
(2) update in-memory BBT
(3) write bad block marker to OOB area of affected block
(4) update flash-based BBT
Note that we retain the first error encountered in (3) or (4), finish the procedures, and dump the error in the end.

This should handle power cuts gracefully enough. (1) and (2) are mostly harmless (note that (1) will not erase an already-recognized bad block). The OOB and BBT may be "out of sync" if we experience power loss bewteen
(3) and (4), but we can reasonably expect that on next boot, subsequent I/O operations will discover that the block should be marked bad again, thus re-syncing the OOB and BBT.

Note that this is a change from the previous default flash-based BBT behavior. If your system cannot support writing bad block markers to OOB, use the new NAND_BBT_NO_OOB_BBM option (in combination with NAND_BBT_USE_FLASH and NAND_BBT_NO_OOB).

---
There are some small issues that may still be addressed in the future, like checking the OOB before erase/rewrite. This patch still stands as-is, though.

v5: Revise patch description to reflect discussion. Move BUG_ON() to

    nand_scan_tail().

v4: Re-order operations so we write BBM before BBT. This should help with
power cuts. Option for old behavior changed to NAND_BBT_NO_OOB_BBM, use in chip->bbt_options.
v3: Writing to flash-based BBT and to BBM is still default, but
there is a new option NAND_NO_WRITE_OOB that can prevent writing the BBM as well as prevent all other OOB writes.
v2: Explain potential power cut issues and remove option for retaining
old behavior.
v1: Implement option NAND_BBT_WRITE_BBM that causes marker to be written
to both BBT + BBM.

 drivers/mtd/nand/nand_base.c |   46 ++++++++++++++++++++++++++++-------------
 include/linux/mtd/bbm.h      |    5 ++++
 2 files changed, 36 insertions(+), 15 deletions(-)
drivers/mtd/nand/nand_base.c +31 -15
index 2d010e4..5822e3a 100644
@@ -386,15 +386,23 @@ static int nand_block_bad(struct mtd_info *mtd, loff_t ofs, int getchip)
  * @ofs: offset from device start
  *
  * This is the default implementation, which can be overridden by a hardware
- * specific driver.
+ * specific driver. We try operations in the following order, according to our
+ * bbt_options (NAND_BBT_NO_OOB_BBM and NAND_BBT_USE_FLASH):
+ *  (1) erase the affected block, to allow OOB marker to be written cleanly
+ *  (2) update in-memory BBT
+ *  (3) write bad block marker to OOB area of affected block
+ *  (4) update flash-based BBT
+ * Note that we retain the first error encountered in (3) or (4), finish the
+ * procedures, and dump the error in the end.
 */
 static int nand_default_block_markbad(struct mtd_info *mtd, loff_t ofs)
 {
 	struct nand_chip *chip = mtd->priv;
 	uint8_t buf[2] = { 0, 0 };
-	int block, ret, i = 0;
+	int block, res, ret = 0, i = 0;
+	int write_oob = !(chip->bbt_options & NAND_BBT_NO_OOB_BBM);
 
-	if (!(chip->bbt_options & NAND_BBT_USE_FLASH)) {
+	if (write_oob) {
 		struct erase_info einfo;
 
 		/* Attempt erase before marking OOB */
@@ -407,23 +415,17 @@ static int nand_default_block_markbad(struct mtd_info *mtd, loff_t ofs)
 
 	/* Get block number */
 	block = (int)(ofs >> chip->bbt_erase_shift);
+	/* Mark block bad in memory-based BBT */
 	if (chip->bbt)
 		chip->bbt[block >> 2] |= 0x01 << ((block & 0x03) << 1);
 
-	/* Do we have a flash based bad block table? */
-	if (chip->bbt_options & NAND_BBT_USE_FLASH)
-		ret = nand_update_bbt(mtd, ofs);
-	else {
+	/* Write bad block marker to OOB */
+	if (write_oob) {
 		struct mtd_oob_ops ops;
 		loff_t wr_ofs = ofs;
 
 		nand_get_device(chip, mtd, FL_WRITING);
 
-		/*
-		 * Write to first/last page(s) if necessary. If we write to more
-		 * than one location, the first error encountered quits the
-		 * procedure.
-		 */
 		ops.datbuf = NULL;
 		ops.oobbuf = buf;
 		ops.ooboffs = chip->badblockpos;
@@ -435,18 +437,28 @@ static int nand_default_block_markbad(struct mtd_info *mtd, loff_t ofs)
 		}
 		ops.mode = MTD_OPS_PLACE_OOB;
 
+		/* Write to first/last page(s) if necessary */
 		if (chip->bbt_options & NAND_BBT_SCANLASTPAGE)
 			wr_ofs += mtd->erasesize - mtd->writesize;
 		do {
-			ret = nand_do_write_oob(mtd, wr_ofs, &ops);
+			res = nand_do_write_oob(mtd, wr_ofs, &ops);
+			if (!ret)
+				ret = res;
 
 			i++;
 			wr_ofs += mtd->writesize;
-		} while (!ret && (chip->bbt_options & NAND_BBT_SCAN2NDPAGE) &&
-				i < 2);
+		} while ((chip->bbt_options & NAND_BBT_SCAN2NDPAGE) && i < 2);
 
 		nand_release_device(mtd);
 	}
+
+	/* Update flash-based bad block table */
+	if (chip->bbt_options & NAND_BBT_USE_FLASH) {
+		res = nand_update_bbt(mtd, ofs);
+		if (!ret)
+			ret = res;
+	}
+
 	if (!ret)
 		mtd->ecc_stats.badblocks++;
 
@@ -3222,6 +3234,10 @@ int nand_scan_tail(struct mtd_info *mtd)
 	int i;
 	struct nand_chip *chip = mtd->priv;
 
+	/* New bad blocks should be marked in OOB, flash-based BBT, or both */
+	BUG_ON((chip->bbt_options & NAND_BBT_NO_OOB_BBM) &&
+			!(chip->bbt_options & NAND_BBT_USE_FLASH));
+
 	if (!(chip->options & NAND_OWN_BUFFERS))
 		chip->buffers = kmalloc(sizeof(*chip->buffers), GFP_KERNEL);
 	if (!chip->buffers)
include/linux/mtd/bbm.h +5 -0
index c4eec22..650ef35 100644
@@ -112,6 +112,11 @@ struct nand_bbt_descr {
 #define NAND_BBT_USE_FLASH	0x00020000
 /* Do not store flash based bad block table in OOB area; store it in-band */
 #define NAND_BBT_NO_OOB		0x00040000
+/*
+ * Do not write new bad block markers to OOB; useful, e.g., when ECC covers
+ * entire spare area. Must be used with NAND_BBT_USE_FLASH.
+ */
+#define NAND_BBT_NO_OOB_BBM	0x00080000
 
 /*
  * Flag set by nand_create_default_bbt_descr(), marking that the nand_bbt_descr
Signature · 1 line
1.7.5.4
#1 lore raw

This looks good to me.

Quoted 2 lines · Brian Norris

There are some small issues that may still be addressed in the future, like checking the OOB before erase/rewrite.

There may be some reasons for always re-writing the OOB marker (e.g. floating bits after interrupted erase/write operation), but this can be discussed in the future...

Angus

#2 lore raw
In reply to Brian Norris
Quoted 5 lines · Brian Norris

Currently, the flash-based BBT implementation writes bad block data only to its flash-based table and not to the OOB marker area. Then, as new bad blocks are marked over time, the OOB markers become incomplete and the flash-based table becomes the only source of current bad block information. This becomes an obvious problem when, for example:

Pushed to l2-mtd.git, thanks!

Signature · 2 lines
Best Regards,
Artem Bityutskiy

---------------------------------------------------------------------
Intel Finland Oy
Registered Address: PL 281, 00181 Helsinki
Business Identity Code: 0357606 - 4
Domiciled in Helsinki

This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.